AI as a fact-checker: What US voters expect from chatbots

What this is about

Someone asking a chatbot about politics is not necessarily looking for a voting recommendation. Often, the aim is to compare positions or check a claim they have already encountered. Research published by Pew Research Center on October 23, 2026, makes that distinction visible: Eight percent of US adults say they are using AI chatbots for information about the 2026 elections.

Within that group, 76% seek information about candidates' positions on issues. Another 71% want to check the accuracy of things they have heard or read elsewhere about candidates or election issues. That is the important point: People are using AI not just for information, but as a supposed second layer of verification. The survey did not test whether the answers were actually correct.

Illustrative photograph: A ballot drop box in Boston, photographed in 2022. Photo: Whoisjohngalt, Wikimedia Commons, CC BY-SA 4.0, unchanged.

What the Pew survey actually does

Pew surveyed 5,273 adults from its American Trends Panel between October 12 and 18, 2026. The published analysis describes whether people use chatbots for election information and what they want from them. It is not an examination of individual chat transcripts or a competition between different models.

The denominator matters. The eight percent figure refers to all US adults. The figures for specific purposes refer only to those using chatbots for election information. Within this smaller group, 65% seek information about candidates' lives and careers. Forty percent want suggestions for making their voting choices, while another 40% seek information about registration, polling places or voting by mail. A person can have several purposes, so these percentages must not be added into a single total.

Age also makes a difference: Eleven percent of adults younger than 50 use chatbots for election information, compared with four percent of those aged 50 and older. Pew describes this as a pattern broadly resembling age differences in chatbot use overall.

Why it matters

The findings put limits on two common narratives at once. Chatbots are not yet an election-information tool used by most of the population. Within the user group, however, they are not merely a novelty: People bring questions about political positions, questionable claims and the mechanics of casting a ballot.

For fact-checking in particular, a plausible explanation is different from verifiable evidence. A fluent answer can be useful if it points to an original speech, a policy platform or the responsible authority. It does not replace that source. This practical conclusion is our interpretation, not a quality advantage of particular products measured by Pew.

For newsrooms and developers, this raises a concrete design question: Can readers see where a claim comes from, when the source was published and whether it applies to the right location? The study does not prove that a particular interface produces better decisions. It does show that such verification tasks are among the purposes frequently reported by the surveyed users.

In plain language

Using a chatbot to check facts is like getting help finding something in a library. The helper can point to relevant shelves and summarize a text. That does not yet establish that the selected book answers the question or is up to date.

The Pew survey effectively counts how many visitors seek this help and which questions they bring. It does not check whether the helper then picks the right book off the shelf. Confusing the two turns a usage statistic into an unintended seal of approval.

A practical example

A fictional reader wants to compare two candidates on three topics: healthcare, taxes and energy. They ask a chatbot for an overview containing six supported entries, one per candidate and topic. Each claim should include an original source and its date.

The reader then opens all six links personally. If an entry lacks traceable evidence, that point remains marked as unresolved. A question about their own polling place is checked directly with the responsible election authority as well. The numbers in this example describe only the imagined workflow, not an experiment or a measured success in Pew's study.

Scope and limits

• Purpose is not accuracy: Respondents report what they use chatbots for. This establishes neither the error rate of the answers nor the quality of a fact-check.
• Use is not demonstrated influence: The research does not show whether a chatbot answer changed a voting decision or actually caused someone to cast a different vote.
• The population is limited: The results concern US adults and the 2026 elections. They cannot simply be applied to Germany. Percentages within the smaller user group also do not describe the entire population.

SEO & GEO keywords

Pew Research Center, AI chatbots, US elections 2026, election information, fact-checking, political information, American Trends Panel, media literacy, source verification, voting decisions